Skip to content

Latest commit

 

History

History
255 lines (213 loc) · 7.64 KB

README_EN.md

File metadata and controls

255 lines (213 loc) · 7.64 KB

Dify Workflow Generator Prompt

Overview

This repository provides prompt definitions for generating workflows in Dify. Following detailed prompt specifications written in YAML format, you can create workflows for various use cases.

Features

  • Simple workflow generation using three basic nodes (Start, LLM, End)
  • Support for advanced node types:
    • Question Classifier
    • Knowledge Retrieval
    • HTTP Request
    • JSON Parser
    • IF/ELSE Node
    • Code Node
    • Variable Aggregator Node

File Contents

  • workflow_generator_prompt.yml
    • Definition of workflow generation prompts
  • dify_chatbot/DifyWorkflowGeneate.yml
    • Dify chatbot with workflow_generator_prompt.yml applied
  • example/manual_search.yml
    • Example of workflow generation

Usage

  1. Clone the repository:
https://github.com/Tomatio13/DifyWorkFlowGenerator.git
  1. Import the workflow generation prompt dify_chatbot/DifyWorkflowGeneate.yml into Dify. After importing, you will see a screen like this: Dify Workflow Generator

Note: Please run this prompt with claude-3-5-sonnet, not gpt-4o. Workflows will not be generated correctly with gpt-4o.

  1. Prepare necessary information:

    • Purpose of the workflow
    • If using knowledge control blocks, look up the dataset_ids:
      • How to find them: Create a workflow that includes knowledge. Export the DSL of the created workflow. 
        type: knowledge-retrieval
title: Ubuntu Knowledge Retrieval
dataset_ids:
    - 84782981-6a4d-4d19-9189-dd72fe435a57
retrieval_mode: multiple
        Note down the dataset_ids as shown above.

  2. Create and execute the prompt as shown in the image below: Dify Workflow Generator A YAML file that can be imported into Dify will be generated as shown in the figure.

  3. Copy the generated YAML to a separate file and import it into Dify.

  4. Below is a screen showing example/manual_search.yml after import: Dify Workflow Generator

Sample

example/adviser_bot.yml

  • Workflow utilizing question classifier, knowledge retrieval, IF/ELSE branching, template transformation, and parameter extraction Workflow Image

Supported Node Types

Basic Nodes

  • Start Node: Accepts user input
    • Text input (short text/paragraph)
    • Numeric input
    • File input (documents, images, audio, video)
  • LLM Node: Processing using OpenAI GPT-4
  • End Node: Output results

Extended Nodes

  • HTTP Request Node: External API integration
  • JSON Parse Node: JSON processing
  • Question Classifier Node: Input classification
  • Knowledge Retrieval Node: Information search from datasets
  • IF/ELSE Node: Conditional branching
  • Code Node: Python code execution
  • Variable Aggregator Node: Aggregates output from multiple nodes
  • Document Extractor Node: Text extraction from documents
  • Template Transform Node: Template-based string generation
  • Answer Node: Response output control
  • Parameter Extractor Node: Parameter extraction from text
  • YouTube Transcript Node: YouTube video subtitle retrieval
  • JinaReader Node: Web page content extraction
  • TavilySearch Node: Web search execution

Workflow Examples

Simple Processing Flow

Start → LLM → End

Flow with Branching

Start → Question Classifier → Knowledge Retrieval → LLM → End
                          → Knowledge Retrieval → LLM → End

IF/ELSE Branching Flow

Start → IF/ELSE → LLM(True) → End
               → LLM(False) → End

Variable Aggregation Flow

Start → LLM1 → Variable Aggregator → End
      → LLM2 ↗

Document Processing Flow

Start(File Input) → Document Extractor → LLM → End

Template Transformation Flow

Start → Template Transform → LLM → End

Complex Processing Flow

Start(File) → Document Extractor → Template Transform → LLM → End

Response Control Flow

Start → LLM → Answer → End

Complex Response Flow

Start → LLM1 → Template Transform → Answer → End
      → LLM2 ↗

Parameter Extraction Flow

Start → Parameter Extractor → LLM → End

Complex Parameter Processing Flow

Start → Parameter Extractor → Template Transform → Answer → End

YouTube Transcript Flow

Start → YouTube Transcript → LLM → End

Web Content Processing Flow

Start → JinaReader → LLM → End

Web Search Flow

Start → TavilySearch → LLM → End

Complex Media Processing Flow

Start → YouTube Transcript → Template Transform → LLM → End
      → JinaReader → ↗

Limitations

  • Only supports OpenAI's gpt-4o model (change model settings on the Dify screen)
  • Boolean type not available in code nodes (use 0/1 with number type)
  • File input supports only the following formats:
    • Documents (PDF, Word, etc.)
    • Images (JPG, PNG, etc.)
    • Audio (MP3, WAV, etc.)
    • Video (MP4, etc.)
  • Template Transform Node constraints:
    • Output is always string type
    • Only supports Jinja2 template syntax
    • Complex control syntax not recommended
  • Answer Node constraints:
    • arrayObject type variables not supported
    • Special behavior in chat mode
    • Variable references only in {{#nodeID.variableName#}} format
  • Parameter Extractor Node constraints:
    • Parameter names must be unique
    • Required parameters must be extracted
    • Image processing only available with supported models
    • Inference mode limited to prompt or function_call
  • Unsupported nodes:
    • Iteration nodes
    • Variable assignment nodes
    • List processing
    • Tool nodes in general
  • YouTube Transcript Node constraints:
    • Only supports public videos
    • Only works with videos that have subtitles
    • Requires subtitles in specified language
  • JinaReader Node constraints:
    • Only supports accessible web pages
    • JavaScript dynamic content may not be fully retrieved
    • May require proxy settings
  • TavilySearch Node constraints:
    • Limited number of search results (max 10)
    • Possible access restrictions for some domains
    • Processing time varies with search depth

About difyDslGenCheck.py

Note: This is an experimental feature and may often fail to generate workflows. Please use Dify's workflow generation feature instead.

Features

  • Generates DSL files based on workflow overview
  • Checks if the generated DSL file structure is correct
  • Modifies DSL files based on check results

Usage

  1. Describe the overview of the workflow you want to generate in difyDslGenCheck.py:
    wanted_workflow = """
    Prompt:
    Purpose: A tool to research destination information like a travel concierge
    1. Get input of destination name as input information.
    2. Search the internet and obtain 3 URLs containing tourist and gourmet information about the destination.
    3. Retrieve information from the 3 URLs. Process the 3 URLs in parallel.
    4. Input the information obtained from the 3 URLs to LLM, organize the destination information, and output it. The output should be formatted as an article suitable for publication in a travel magazine.
    """
  1. Execute the script:
export ANTHROPIC_API_KEY="your_anthropic_api_key"
python -m venv dify_env
source dify_env/bin/activate
pip install -r requirements.txt
python difyDslGenCheck_en.py

License

MIT License

Contributing

Pull requests and issue reports are welcome.

Related Links